A novel application of a technique developed from chaos theory is used in describing seasonal to interannual climate predictability over the Northern Hemisphere (NH). The technique is based on an empirical forecast scheme - local approximation in a reconstructed phase space - for time-series data. Data are monthly 500 hPa heights on a latitude-longitude grid covering the NH from 20° N to the equator. Predictability is estimated based on the linear correlation between actual and predicted heights averaged over a forecast range of one- to twelve.month lead. The method is capable of extracting the major climate signals on this time scale including ENSO and the North Atlantic Oscillation.